[Cosmology Seminar] Cosmological Inference with Machine Learning

Simulation-based methods have the potential to give access to cosmological information that is out of reach for traditional perturbative, explicit-likelihood methods. In order to realize this promise, however, the related problems of computational feasibility and sensitivity to systematics need to be addressed. I will give an introduction to simulation-based inference with a focus on the late-time large-scale structure and then present work on multi-fidelity training and physics-guided data compression.